Time-Series Laplacian Semi-Supervised Learning for Indoor Localization ^†.

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Machine learning for indoor localization traditionally requires large, labeled datasets, hindering practical application.
• Semi-supervised learning offers a solution by reducing the need for labeled training data.
• Existing semi-supervised methods face challenges in efficiency and accuracy for real-world indoor positioning.

Purpose of the Study:

• To propose a novel time-series semi-supervised learning algorithm for enhanced indoor localization.
• To leverage unlabeled data to improve the accuracy and efficiency of smartphone user positioning.
• To address limitations of conventional semi-supervised approaches in indoor localization tasks.

Main Methods:

• A new time-series semi-supervised learning algorithm is developed, utilizing spatio-temporal relationships in unlabeled data.
• Pseudo-labels are generated from unlabeled data to augment limited labeled training sets.
• A balancing-optimization learning algorithm constructs the final positioning model, incorporating Wi-Fi Received Signal Strength Indicator (RSSI) measurements.

Main Results:

• The proposed algorithm demonstrates superior performance compared to existing semi-supervised methods.
• Performance improvements are observed across varying numbers of training data points and access points.
• Analysis of learning parameters provides insights into the method's enhanced performance.

Conclusions:

• The developed time-series semi-supervised learning algorithm effectively improves indoor localization accuracy and efficiency.
• The method's ability to utilize unlabeled data offers a significant advantage over traditional approaches.
• An extended scheme incorporating particle filters and floor plan data further enhances localization capabilities.